Michael Lynch, CEO of Praxie and former head of Internet of Things at SAP, joined Jimmy Carroll, vice president of operations at TECH B2B Marketing, on the Manufacturing Matters podcast, where they discussed the benefits that the convergence of artificial intelligence (AI) and process digitalization can bring to the manufacturing industry.

Praxie’s enterprise software solution uses AI to help businesses digitally transform their operations at low costs. The software offers preconfigured applications, or users can work with Praxie experts to create customized solutions.

During the podcast, Lynch discussed major trends in AI and manufacturing and how manufacturers should think about and use the technology. He also outlined barriers to implementation and basic steps to get started with AI.

Jimmy Carroll: [00:00:07] Hi everyone. My name is Jimmy Carroll. I’m the vice president of operations at Tech B2B Marketing, and I’m joined today on the Manufacturing Matters podcast by Michael Lynch, who is the CEO of Praxie. Michael, thanks, first of all, for joining us today of course. And I’d like to first start by asking you a little bit about your background and ultimately how you ended up co-founding Praxie.

Michael Lynch: [00:00:29] Yeah, I have one of the weirdest backgrounds ever for being in high tech. I started my career as an actor on Broadway. I used to have a lovely mullet and did “Les Miserables” and lots of shows and then got involved in making computer games, which is a natural transition. And we were able to take the company public. And so that taught me the whole startup thing. From there, because of the 3D aspects, I got involved in 3D CAD, and from there we built some applications and platforms, a company called Right Hemisphere, which we then sold to SAP. And then I ended up running the Internet of Things division for SAP for a couple of years prior to starting Praxie. So I learned a lot about that, Industry 4.0, the whole background that SAP had in it. So that’s the brief background, from singing and dancing to singing and dancing about tech.

Jimmy Carroll: [00:01:21] So Praxie, so the website, anyone who goes there, it says the company offers AI-powered digital transformation software. So AI and digital transformation both are sort of industry buzzwords, right? But they mean real things. And they do real things of course. So what exactly does this mean? I’ll start there.

Michael Lynch: [00:01:41] Well, if I start back at kind of why we founded the company, when I was at SAP, and every manufacturing company I went to had a whiteboard and huddle boards and was trying to do lean processes and just a mountain of Excel, even if they had SAP and good MES software, etc. So if you’re going to drive digitization and you’re going to use AI, you can’t do it when you’re in paper or even what I call digital paper, Excel. You have to get the data digitized so you can use AI in it. So we thought the first job to support was how do you create a platform that’s as easy to digitize your processes, whether those be lean processes or scheduling or whatever you want to do, tailored to your company? And how do you do that in a way that takes days versus months or plus-years to digitize these processes? So we built a platform to do that. We then allowed it to connect to data sources, and then we imbued it with AI workflows so that we can take all of the goodness of the applications that exist, surround that with a digitization platform that can then use AI to drive automation and increase productivity of the workers at their point of work.

Jimmy Carroll: [00:02:52] Okay, so with AI, I’m going to get into this a little bit more. But AI means a lot of different things to a lot of different people. And I will ask you about that more. But how does your platform leverage AI specifically?

Michael Lynch: [00:03:07] Well, traditionally, as you know, you guys are experts in this, the previous generation of AI that we had at SAP was looking at vibration analysis and trying to do predictive maintenance and all the kinds of things that we’re familiar with. With the large language models, you have a completely different analysis capability than we had before and the ability to use that for people. So what we did is we built AI into the fundamental workflow engine of our system, and that allows us to take processes like gemba walk. Let’s just do something super simple, like a gemba walk or a 5S audit. And you can, as a manager, now you’ve got an application to do this super easy. You go and you do the work. And then the AI, when you save the data, will go create summaries of everything that you’ve done, fairly useful, but then it will also go out if you want and suggest how you might address some of those issues based on the best practices that they can find in the internet. And then it can also generate a list of actions that you can then assign to your team. So what we’re trying to do is make the use of AI very, very practical and simple and tie into processes that people already understand versus making the box gigantic. There’s some really good use cases, as you and I were talking prior to the start around vision and using vision for quality of manufacturing or for the quality of the parts. But there’s a huge amount of areas where AI can be used for improving the workers’ productivity and giving them a buddy that makes them four times smarter than they already are at whatever they’re trying to do.

Jimmy Carroll: [00:04:41] That’s interesting. And I want to touch on that vision topic a little bit later too. But staying on the topic of AI, so it sounds like you’re using some generative AI in this process, is that right?

Michael Lynch: [00:04:55] Yeah, we’re using large language models, and we’ve been toying with different ones. For us I think the models are being somewhat commoditized in their capabilities. And so the real question is what you do with them. So we’re always testing different models as they come out. But the generative AI on how we can use it to create effectively that copilot at every aspect of the work is where we’re focused.

Jimmy Carroll: [00:05:22] It’s cool. I mean, I’ve seen on the show floors, at any of the large automation shows, I’ve seen some actual applications of generative AI where it’s adding value to the manufacturers and not just generating cool pictures or rewriting the end of “Game of Thrones” for people online for fun.

Michael Lynch: [00:05:44] That’s cool too, but it won’t help you get your product out.

Jimmy Carroll: [00:05:47] Okay, so there’s a lot of different apps and segments on your website in terms of how your software can help you. What are some of the most common ways, like how has it and how can it help manufacturers? When people reach out to you, what are the most common things they are coming to you looking for you to solve?

Michael Lynch: [00:06:08] So I’ll talk about two of them. The most common is the simple huddle board concept, that even if you go to the most advanced manufacturers around, they tend to still be on paper. You can’t use AI, as I said, if you’re there. So what’s unique about the way we do it is that we have lots and lots of templates. I mean, probably the largest group of template processes on the internet for manufacturing and others. But think of those as frameworks that we tailor to the customers. So because every manufacturing company I’ve been in runs those huddle boards and their lean processes differently, they have the same principles, but the tools have been modified to fit their needs. So we can digitize those super, super quickly and give them a complete digital capability that you can use by line, in area, shop, multi-plant, etc. and roll all the data up. So that’s the simplest way to get them started in the digital realm in complement to what they’re already likely doing on the line, where they’re taking machine data and running how many parts, how much waste, etc. The second thing that we’re doing is sitting on top of those areas. That might be IoT data that’s coming in. It might be data that’s sitting in an SQL database that we’re probing constantly and running applications on top of those that notify people, that track all the action items, that run up to their digital KPI or command centers or obeya rooms. So those are the two. One is just going really easy and low, digitizing the basic stuff that’s in paper. And the second one is sitting on top of all the information that they have and generating applications that keep everybody informed and on action and on task.

Jimmy Carroll: [00:07:45] So when we’re discussing topics like automation on this podcast, generally we are talking about robotics and machine vision and motion control and AI. But what you’re doing is talking about automation in manufacturing. You touched on it briefly, but what are some ways that digital transformation software like yours can augment or work together with these technologies, these technologies being vision. For example, a machine vision system’s capturing images on a line. Can your software leverage this data for things like process monitoring, quality assurance. You mentioned predictive maintenance. Can it do that?

Michael Lynch: [00:08:31] Yeah. So what we’ve been doing, just working with a process manufacturer recently that the machine vision is actually tracking the bottles that are coming out of the bottle blowing machine and tracking the quality of those. And they’re getting too much scrap. And so what you can do is we can create an application that takes all that information and from that monitoring can generate a process to notify, which is great, and a lot of applications can do that. But what we also do is we take the best practices, and this gets to at the core what you’re trying to do with automation is how do we do more with the people we have? Because that’s, as you know, what the main issue is. I’ve never been in a plant where they say, “Oh, we have so many extra people. It’s great.” So what we’re doing is taking that error and that issue and through our AI and training applications looking at the data that the company already has, as well as work instructions and things that are either built or that we build for the client.

Michael Lynch: [00:09:34] And not only do you get as the operator, instead of just saying, “Oh I’m having too much scrap. Push the button, I need to talk with maintenance,” We provide at the point of work, a bunch of AI-driven things that that person can do, and they can run through a routine based on all of these things and use that as the step one before they go call maintenance. And so through those kinds of applications where, as I mentioned, you’re sort of trying to make every worker in the place, from the line worker up to the executives, much smarter by using AI as a copilot to what they’re trying to do. So in that case, you’re reducing the amount of downtime and the amount of times the maintenance person gets billed and you’re increasing the amount of bottles in that case that get made. And so both of those have huge implications on the ROI, on implementing a software like ours.

Jimmy Carroll: [00:10:27] Okay, so one of the things that you touched on that’s interesting to me is joking about maybe not every place has enough. It’s not a joke, of course, but there’s a labor shortage, there’s a persistent labor shortage. And again, when we get on this podcast or in conversations I have at trade shows or with whoever, automation has become increasingly important over the last several years. COVID has highlighted this, a labor shortage, political tensions, whatever. With your product or just digital transformation software in general, what are some ways that this can help with the labor shortage?

Michael Lynch: [00:11:08] Well digitization at its core, nondigital work takes a lot more time. That’s why the software is eating the world, as Marc Andreessen says. So the very basic idea is that with a digital environment, whether that be an MES system or an ERP system, that your productivity dramatically improves. And if you think about automation the way that it is talked about today, we’re thinking about robotics and machine vision and those kinds of things, but there’s just a tremendous amount of low-hanging fruit and manufacturers that they can actually adopt very, very rapidly. And so the base and core value proposition is the same reason that we’re using this software instead of meeting, because digital improves productivity. And in this case you’re not flying places. But in the case of managing all of the lean processes in manufacturing, we talked about gemba walk and these other kinds of things, that’s just taking processes that would normally be scribbled down on paper, that you’d have to have somebody then go say, well, what are the action items, that you then try to figure out and you lose productivity because you’ve lost the list of actions that happen. And so there are applications for gemba walks, but they’re not applications for gemba walks that do exactly what that manufacturer wants and 700 other things they might need.

Michael Lynch: [00:12:27] And so what we’re trying to do is go after that low-hanging fruit of just get digital. So you get the productivity gain and the automation that software gives you on everything. Then once you’re there, you can then layer on AI. You can layer on connectivity to real-time data streams, whether that be traditional machine data or whatever or the kinds of things that you might be getting feedback from vision systems or robotic systems or any other kind of system. And you now have a layer that you can turn into any workflow application you want while talking to the simplest thing, which is a form filled out by a floor leader walking around the floor, all the way up to a vision system giving you cues about a particular thing you’re interested in. And we can provide that unified platform that surrounds everything you’ve got. So at some level what we’re talking about is getting very basic about the stuff that’s low-hanging fruit but giving you the overall capability to expand on that and really adopt all these other technologies into an overall digitization platform.

Jimmy Carroll: [00:13:30] It sounds like a lot of these things, all of them kind of tie in to these, again, buzzwords of Internet of Things or Industrial Internet of Things and Industry 4.0. These terms have always been a little bit nebulous to me. I understand what they refer to and what they mean, but . . .

Michael Lynch: [00:13:49] I agree. Can I just jump on that one second? Because I ran the division called the Internet of Things, and I was part of that whole group of software guys who said, “Oh, we are changing the world.” The reality was, as you know, the manufacturing world had been doing the Internet of Things for quite a long time. And as I see the Internet of Things versus digitization, and this is obviously not meant to be a hard line and there’ll be a thousand people with different opinions and they’re probably smarter than I am. But when we were doing the Internet of Things, it was really all about the data and the sensors and getting access to the data and the sensors.

Michael Lynch: [00:14:24] And then the application, what do you do with it? Well, you can buy SAP platform and do something there. You can buy something over there, something over there. But that layer of what do you do with it is in my mind been traditionally left to IT teams or development groups to build those applications on top of the IoT layer. So when we talk about digitization, I’m talking about the entire ability to digitize anything, whether it’s a manual process, it’s a lean best practice that you’re doing in Excel, whether it’s IoT-based stuff, whether it’s database connectivity stuff, so that you can run automated processes. It might be as simple as your defective parts application. How do you do that at your plant today? How do you manage that? My guess is there’s tons of Excel in a lot of these processes. So what we’re trying to do is sit on top of what is traditionally called IoT but also machine vision or these other kinds of things that we can then make applications very rapidly and that people can actually get real value from these technologies versus theoretical value. And that’s exactly what you were saying. There’s a lot of theoretical value in these things. But how do you very rapidly get actual pragmatic value out of them? And we’re trying to create a layer in which applications can be built very rapidly for that.

Jimmy Carroll: [00:15:42] So that’s interesting. So both in terms of what what Praxie does and in general, like these terms, IoT, Industry 4.0. You know, a lot of larger companies they’re already doing this. But maybe some smaller, medium-sized manufacturers who are maybe behind the curve on the technology adoption. What are the most fundamentally important aspects of adopting these approaches? What do they need to be doing? Like what’s the hit list for them to make sure that they’re not losing data? Do you know what I’m saying? How do they get started?

Michael Lynch: [00:16:23] Yeah. Let me address the large companies as well. The way that they’ve done it. And I would make a bet there’s still a huge amount of Excel running around in those organizations as well. But the way they’ve done it, and it makes a lot of sense, is they have the IT teams and infrastructures and custom development to get it done. If you’re at a major manufacturer in the United States or Japan or anywhere else, you have massive amounts of homegrown applications that are surrounding your off-the-shelf applications. For smaller and midsize companies, that’s not an option. They don’t have that. So they rely on documents in Excel. And so our recommendation to the bigger ones is you can replace all that custom development with something like us. And we’re not the only people in the world who do this concept. But I think our idea was: How do we make this as easy as making PowerPoint? And we’ve gotten pretty close. And so we can build applications in days. But if you’re a smaller to mid, to answer your question about how do you get started, I think it has two aspects. One is: Where is there a lot of paper in Excel in critical processes? And try to bucket that into something that you can take on at one point. Meaning if we just solve this parts issue problem, it will have a massive impact on our bottom line, or scrap or whatever, and take that one problem and digitize what is being done around it.

Michael Lynch: [00:17:48] And as I said, it’s sitting in Excel or papers or other things like that. Digitize that and keep it very contained to that and the ROI. Get a success, and then move on. But the critical thing, and again I’m pitching Praxie here. The critical thing is do you have a strategy? And I’ve heard other people talk about this as well. And I think it’s also important at an infrastructural layer, because you need to be able to swap out the pieces and parts if any one of them don’t work. But for our vision, Praxie represents a strategy that I can go after my defective parts area. I can build a bespoke application in a matter of days or weeks and be up and running and get massive ROI and get the payback on this in a few months. And then I can go, “Oh, I can go take on the next thing.” And so you limit the risk, you maximize the value, you put points on the board as a plant manager or an IT manager who’s trying to do that. And you containerize it into an area that has high value but isn’t exposing your entire infrastructure to risk. And then once you get that developed, then you see the value and then you take on the next and the next and the next and the next.

Jimmy Carroll: [00:18:54] Well thanks for that. I don’t mean to keep steering the conversation toward AI, but when people think of AI, it means a lot of different things to different people: ChatGPT, in the manufacturing space it means machine learning, deep learning. Either way, though, there’s a lot of hype about AI. We see AI commercials on football games, right? What are some of the major trends in AI, both in terms of what you guys do in AI in the general space as an augmenting tool to machine vision, for example?

Michael Lynch: [00:19:32] Yeah, that’s a really interesting question. I think the most interesting thing that, from a trend standpoint, you just saw Elon Musk came out with his version I just saw. I didn’t read the article, but I just saw he came out with one yesterday or something. I think there’s going to be a trend to commoditization of large language models and what that means to me for your viewers is: don’t get too locked in on which model. Build a strategy that allows you to determine which model may be the best and/or cheapest model over time. I think all of them will somewhat race to the bottom, sort of like AWS and Microsoft and Google are doing with the overall platform. I think there’ll be an absolute easy way to get the AI structures, the best language models out of those. But there’s going to be a trend toward commoditization. Commoditization and as it goes beyond that into the true general intelligence, none of us know what’s happening at that. But the thing that I think is most important is to find the very pragmatic ways that you can use AI.

Michael Lynch: [00:20:41] For instance, when you’re talking about computer vision and you’re looking, you can use these models to find errors in your parts, errors in your manufacturing. You know, you’ve probably done podcasts on these things. Tremendous opportunities there. I think the one that we’re focused on, which I think requires a lot of work on the side of the software vendor or the IT team that’s building things out, is to have a copilot that really helps to automate a lot of the thinking, if you will. It’s not the doing. There’s studies that say, I don’t have the source in front of me, but the average AI is two to three times more creative than the average individual on finding solutions for particular problems. So if you think of AI as your copilot buddy and all of these processes, it’s a very pragmatic way to embed AI into every aspect of your company to help people be more productive. I think that’s the short-term biggest opportunity that I’ve seen around how to use AI in its current forms.

Jimmy Carroll: [00:21:44] So kind of on the flip side of that, what are some of the barriers to using AI? And then for those who aren’t already, how does one go about getting started when it comes to adopting AI, again, both in terms of what you do and in general?

Michael Lynch: [00:21:57] I think the biggest barrier is understanding the API prompt structure. And I don’t even know if how many guys on here have been working with that. But basically each of the APIs have a prompt structure, which is asking the AI questions. You create a cascade of prompts. There’s lots of different techniques in there to ground and try to make things not hallucinate. If you’re not familiar with those terms, the AIs make things up if they don’t have a good answer. So basically that’s the biggest barrier, because 90% of the world just goes to a ChatGPT or Bard or something like that and asks a question. And so you can ask a cascade of questions, but you’re in this chat-like environment, which is incredibly useful, but it’s not a business environment. You’re having a conversation about something, so in opposite of a business application that’s running you through a particular process, let’s say an A3, and you say here’s my problem: My refrigeration is having issues and it’s spoiling my food. And you lay this problem out. You say, help me. The AI goes out and finds you 20 problems that might be causing that. And you go, wow, I think problem number three might be it in my case. You say, what should I do about it? The AI goes, here’s six things you can do to solve that problem. There was no person who knew that stuff in that manufacturing plant and that AI just helped that plant manager dramatically at the point of work. And to make that happen is 50 different prompt infrastructures that are under the hood. You could go to ChatGPT and spend an hour and a half trying to figure that out and maybe get something. But I think the barrier is understanding that. And I would say everybody should go find different ways to use ChatGPT. I think Microsoft and Apple and other vendors are coming out with good structures. In our case, we can embed an AI flow into almost any process to help the worker. And that’s really where a platform that’s a general digitization platform says, What is the business problem? What do you need to do? How quickly can we get it growing? And then how can we help your workers be more effective in that process? That’s a very clean and focused way to get real value out of AI today.

Jimmy Carroll: [00:24:15] Yeah. So like in terms of getting value out of it, that’s something I wanted to ask about. Maybe a hard question, but what should the ROI be in terms of AI adoption?

Michael Lynch: [00:24:28] I would not put AI into any different category than any technology adoption. I mean, if you’re going to spend $1 million on a new manufacturing line, the assumption on all projects is, is an IRR associated with it? And so the thumb in the air that I’ve heard a lot from customers is they like less than one year ROI on the cost of implementing something. We try to do it in two to three months. On very focused applications, we try to get the ROI for the client in that amount of time. But I think AI is the same. If you’re going to take an AI vision system and you’re going to implement that for improving any sort of process, I would assume that they’re looking for the same less than a year ROI on that investment.

Jimmy Carroll: [00:25:15] Interesting. Okay, Michael, I’ve asked a lot of the questions that I intended to ask, but kind of a general one: What are some of the other technology trends in manufacturing that you’re most excited about now and going into the future?

Michael Lynch: [00:25:33] That’s a really interesting question. I think that the trends that I’m excited about . . . there’s so many. So in my previous company we used to do 3D, and in this company I’m not doing 3D. I think 3D vision and those kinds of augmented reality things are exciting. But my honest opinion is if you look at the latest work being done, we were doing systems 10 years ago that are clunky but you could use for work. And I don’t think they’ve improved dramatically. I think we’re still a decade or so away in that. I think the areas of most opportunity are the areas that I’m talking about. But if I was going to talk about things that are not in the Praxie realm, I think the vision systems are really, really interesting, the robotic systems, but only in the context of onshoring. Because if you look at the North American economy, this is sort of beyond technology. But if you look at the North American economy and the value of onshoring because of the COVID supply chain issues and the political instability globally that’s happening. So to me, the most exciting thing is how is the next generation of factories being set up or retrofitted back in the United States going to be different and more advanced than the ones that were built 20 or 30 years ago or 50 years ago that got offshored. I think that combination of IoT capability built in at the core with application development and digitization platforms like Praxie, with robotic vision, with vision systems and robotics, that next gen of factory and how that relates to the humans in it, is I think the most exciting trend. It’s the onshoring on top of all that that I find most interesting.

Jimmy Carroll: [00:27:22] Yeah, that’s really cool. It’s interesting. I think a lot of what you said, I think of some of the trends specifically in industrial automation, things like the rise in the popularity of application-specific systems that make it easier to deploy specific task-related systems for, whatever, bin picking or machine tending or things like that. They’re kind of lowering the barrier to adoption for automation. And AI, both in terms of machine learning, deep learning software that can augment a traditional machine vision approach, and with what you’re doing and is ease of use. It’s becoming easier to use. No-code, low-code options. So there are these trends that are making automation easier. And automation was always ripe for growth, COVID or no COVID, in my opinion. Like I said earlier, some of these medium-size, smaller-size manufacturers start to pick it up as different parts of the country, parts of the world rather, that have maybe slowed on adoption start to pick it up and use it for different things. These systems can do more. They’re easier to use. They become more important.

Michael Lynch: [00:28:41] I would say one other thing about adoption. One of the things that we’ve really focused on is, you mentioned that systems, verticalized systems are easier to use. They’re more simplistic. And that’s good. One of the things that we’ve focused on is the ability to digitize what the worker is already using, meaning that generally they have forms and processes. And so we turn those into digital forms and processes that the worker doesn’t really need to be trained on because they know how to do it. They’re just doing it in a screen versus on a pad. And so that adoption curve is incredibly important to getting the ROI in regardless of the process. So I think that’s beyond simple verticalized solutions. Simple non-changing of your analog to digital is another strategy to get people to adopt much, much quicker.

Jimmy Carroll: [00:29:28] Yeah. Interesting. Michael, I don’t have any other questions for you. Is there anything else that we haven’t talked about that you feel like we should mention that I haven’t touched on?

Michael Lynch: [00:29:40] You know, I’d love to talk with any of the folks out there, happy to build the community. And I really appreciate it. It’s super fun.

Jimmy Carroll: [00:29:48] And if they want to learn more about Praxie, they can just go to praxie.com. I believe that’s the right URL.

Michael Lynch: [00:29:53] Praxie.com, and you can set up a meeting with one of our team there. And we’re happy to learn about what you’re trying to achieve. And we can get you up and running quickly.

Jimmy Carroll: [00:29:59] Well I appreciate that Michael. Everyone, thanks so much for joining. This has been the Manufacturing Matters podcast. It’s manufacturing-matters.com. If anybody has questions or would like us to put you in touch with Michael or anyone from the team, would like to join the podcast or anything like that, feel free to reach out to us and otherwise thanks very much for joining us today.